[Ben0000]

Hello. I'm trying to analyse the flow over a semi-cylindrical structure. I have found that on the windward edge the pressure varies wildly between neighbouring cells, with the pressure increasing and decreasing rapidly in a non-physical manner. I was wondering what would be the cause of this error, and how I would go about fixing this. I'm using the LES Smagorinsky turbulence model, if that is relevant. Some screenshots are attached. Thanks.

[Yann]

Hello,

Have you meshed your case with snappyHexMesh? How does your STL look like?

Yann

[Ben0000]

Hi Yann,
Thanks for the quick response. I have meshed this case using snappyHexMesh. I believe my stl is of a decent quality but perhaps not, I've attached it to this reply (I've just changed the file type to .txt so I could attach it to this reply directly). Below is my snappyHexMeshDict:

FoamFile
{
    format      ascii;
    class       dictionary;
    object      snappyHexMeshDict;
}

castellatedMesh on;
snap            on;
addLayers       off;
mergeTolerance     1e-6;

geometry
{
    domain
    {
        type closedTriSurfaceMesh;
        file "Domain.stl";
        regions
        {
            inlet            {name inlet; }
            outlet            {name outlet; }
            free            {name free; }
            floor            {name floor; }
            buildingWall    {name buildingWall; }
        }
    }
};

castellatedMeshControls
{
    maxGlobalCells    20000000;
    maxLocalCells    2000000;
    minRefinementCells    0;
    resolveFeatureAngle    15;
    nCellsBetweenLevels    5;
    allowFreeStandingZoneFaces    false;
    locationInMesh (0.123 0.123 0.123);
    
    features
    (
        {
           file "Domain.eMesh";
           level 0;
        }
    );

    refinementSurfaces
    {
        domain
        {
            level (0 0);
            regions
            {
                floor                { level (0 3); patchInfo { type wall; } }
                buildingWall        { level (3 3); patchInfo { type wall; } }
                free                { level (0 0); patchInfo { type patch; } }
                inlet                { level (0 0); patchInfo { type patch; } }
                outlet                { level (0 0); patchInfo { type patch; } }
            }
        }
    }
}

snapControls
{
    nSmoothPatch 5;
    tolerance 2.0;
    nSolveIter 500;
    nRelaxIter 15;
    nFeatureSnapIter 50;
    implicitFeatureSnap false;
    explicitFeatureSnap true;
    multiRegionFeatureSnap false;
}

addLayersControls
{
    layers
    {
    }

    relativeSizes       true;
    expansionRatio      1.2;
    finalLayerThickness 0.5;
    minThickness        1e-3;
    featureAngle         30;
    nGrow                1;
    maxFaceThicknessRatio    0.5;
    nBufferCellsNoExtrude    0;
    nLayerIter            50;
}

meshQualityControls
{
    maxNonOrtho                65;
    maxBoundarySkewness     20;
    maxInternalSkewness        4;
    maxConcave                80;
    minFlatness             -1;
    minVol                    1e-13;
    minTetQuality            1e-15;
    minArea                    -1;
    minTwist                0.02;
    minDeterminant             0.001;
    minFaceWeight             0.05;
    minVolRatio                0.01;
    minTriangleTwist        -1;
    nSmoothScale            4;
    errorReduction            0.75;
}

writeFlags
(
);

[Yann]

Hello Ben,

Thanks for sharing your files.
The triangles on your STL are pretty elongated. Can you remesh your STL with smaller triangles?
Something like the right cylinder part on this image: http://dawningresearch.org/img2/crea...aorticroot.png

I cannot guarantee it, but I remember a Tips & Tricks session at an OpenFOAM conference where this exact problem was mentioned: elongated triangles on a curved STL surface caused weird pressure jumps even if the OpenFOAM mesh looked OK.

Hope this helps,
Yann

[Ben0000]

Hi Yann,


Thanks for the suggestion. Unfortunately, I've remeshed and rerun the analysis with a refined stl file with no elongated triangles and the results have come back pretty much the same. However, I've done some more investigating and found that the values of nut vary in the same manner as pressure, changing rapidly along the surface (see image nut1). I then looked at how nut varies along the floor, and I was surprised to see that it changes drastically whenever the mesh size changes (see image nut2). Looking back at the building, it's clear that the varying nut occurs when there are changes in the size of the mesh elements (see image nut3). I'm assuming these changes are what's driving the varying pressure distribution. Would you (or anyone else) know why nut changes so drastically with mesh size? Is this an error or am I misinterpreting something? Below is a copy of my 0/nut file.

FoamFile
{
    format      binary;
    class       volScalarField;
    location    "0";
    object      nut;
}

dimensions      [0 2 -1 0 0 0 0];

internalField   uniform 0;

boundaryField
{
    buildingWall
    {
        type            nutkWallFunction;
        value           uniform 0;
    }
    inlet
    {
        type            calculated;
        value           uniform 0;
    }
    outlet
    {
        type            calculated;
        value           uniform 0;
    }
    free
    {
        type            calculated;
        value           uniform 0;
    }
    floor
    {
        type            nutkWallFunction;
        value           uniform 0;
    }
}

[hjasak]

Your surface is not smooth. Calculate the curvature and plot it - it will show you what’s going on.

With a smooth mesh it will be fine

[Ben0000]

Hi Hrvoje,


Could you elaborate on how I would calculate curvature? And how I would generate a smooth mesh? Is snappyHexMesh able to do this or would I need some other tool? Thanks.

[arjun]

Originally Posted by Ben0000

I'm using the LES Smagorinsky turbulence model, if that is relevant. .

This is most probably relevant as you can see that viscosity in this case is control volume dependent and you can see the jumps. The pressure follows the viscosity as you noted.



Try the calculation without LES model and if these jumps disappear then it is the reason why this is happening.

[Ben0000]

Hi Arjun,
I did actually try this model earlier using RANS as the turbulence model instead of LES. It gave me quite similar results and the jumps in pressure were still there, although they were slightly less pronounced due to the "smearing" RANS does. I haven't run the model without a turbulence model yet, and I agree with you that this would probably solve the problem. Unfortunately, the flow is most definitely turbulent given my wind speed, so a laminar analysis would be quite inaccurate. I was hoping to fix this issue in general too, not just for this particular case.
I think at this point the best way to fix this issue would be with a better quality mesh. I've tried changing the minVolumeRatio parameter in snappyHexMesh dict, but I still get these sudden cell volume changes along the surface of the obstacle. Perhaps an external meshing tool like Salome is necessary.

[Alczem]

I would give cfMesh a try too, if you think the mesh is the culprit here. Your geometry looks simple enough that it won't take long to generate a decent mesh.

[arjun]

Well the idea of laminar flow is to see if turbulent viscosity is causing it.

What you describe about the behaviour of turbulence models here makes sense because in LES the turbulent viscosity is function of velocity gradients and volume and in RANS models while control volume does not play direct role, the flow gradients certainly decide the turbulent viscosity.

Now flow gradients do depend on the control volumes and aspect ration and skew etc. So I believe this is what is causing the jump in turbulent viscosity and correspondingly jumps in pressure.

In the solvers that are Rhie and Chow based, many times this variation is reflected in momentum coefficient used in flux calculation and this causes this type of behaviour (Still can't say yet if this is really causing here but mostly seen this case cause in other simulations based on experience).

Originally Posted by Ben0000

Hi Arjun,
I did actually try this model earlier using RANS as the turbulence model instead of LES. It gave me quite similar results and the jumps in pressure were still there, although they were slightly less pronounced due to the "smearing" RANS does. I haven't run the model without a turbulence model yet, and I agree with you that this would probably solve the problem. Unfortunately, the flow is most definitely turbulent given my wind speed, so a laminar analysis would be quite inaccurate. I was hoping to fix this issue in general too, not just for this particular case.
I think at this point the best way to fix this issue would be with a better quality mesh. I've tried changing the minVolumeRatio parameter in snappyHexMesh dict, but I still get these sudden cell volume changes along the surface of the obstacle. Perhaps an external meshing tool like Salome is necessary.